CN113744016A

CN113744016A - Object recommendation method and device, equipment and storage medium

Info

Publication number: CN113744016A
Application number: CN202011219288.3A
Authority: CN
Inventors: 潘扬; 李山林; 张青青; 毛锐
Original assignee: Beijing Jingdong Century Trading Co Ltd; Beijing Wodong Tianjun Information Technology Co Ltd
Current assignee: Beijing Jingdong Century Trading Co Ltd; Beijing Wodong Tianjun Information Technology Co Ltd
Priority date: 2020-11-04
Filing date: 2020-11-04
Publication date: 2021-12-03
Anticipated expiration: 2040-11-04
Also published as: WO2022095701A1; CN113744016B

Abstract

The embodiment of the application discloses an object recommendation method, which comprises the following steps: cleaning order data to obtain an object set corresponding to the order data; the set of objects includes at least two objects; determining the basic heat of each object in the object set; determining a heat characteristic of each object based on the basic heat of each object; ranking the heat characteristics of all the objects in the object set, and determining the objects corresponding to the heat characteristics in a preset ranking range as a recall set; the recall set includes at least two target objects; sequencing the at least two target objects to generate a sequencing result; and outputting the at least two target objects according to the sequencing result. In addition, the embodiment of the application also discloses an object recommendation device, equipment and a storage medium.

Description

Object recommendation method and device, equipment and storage medium

Technical Field

The embodiment of the application relates to the technical field of computers, and relates to but is not limited to an object recommendation method, an object recommendation device, object recommendation equipment and a storage medium.

Background

In the related art, when recommending commodities in a pay Per Sale (CPS) advertisement, generally recommending the commodities according to sales, wherein the sales is statistics of sales in a past period of time, so that the recommending method has certain time delay and can prevent recommended commodity users from purchasing the recommended commodities; therefore, when recommending a product in a CPS advertisement, the product cannot be recommended accurately.

Disclosure of Invention

In view of this, embodiments of the present application provide an object recommendation method and apparatus, a device, and a storage medium to solve at least one problem in the related art, so as to recommend an object with a hot degree more accurately.

The technical scheme of the embodiment of the application is realized as follows:

in a first aspect, an embodiment of the present application provides an object recommendation method, where the method includes:

cleaning order data to obtain an object set corresponding to the order data; the set of objects includes at least two objects;

determining the basic heat of each object in the object set;

determining a heat characteristic of each object based on the basic heat of each object;

ranking the heat characteristics of all the objects in the object set, and determining the objects corresponding to the heat characteristics in a preset ranking range as a recall set; the recall set includes at least two target objects;

sequencing the at least two target objects to generate a sequencing result;

and outputting the at least two target objects according to the sequencing result.

In a second aspect, an embodiment of the present application provides an object recommendation apparatus, where the apparatus includes: the device comprises an acquisition module, a first determination module, a second determination module, a third determination module, a sorting module and an output module; wherein the content of the first and second substances,

the acquisition module is used for cleaning order data and acquiring an object set corresponding to the order data; the set of objects includes at least two objects;

the first determining module is configured to determine a basic heat of each object in the object set;

the second determination module is configured to determine a heat degree feature of each object based on the basic heat degree of each object;

the third determining module is configured to rank the heat characteristics of the objects in the object set, and determine an object corresponding to the heat characteristic within a preset ranking range as a recall set; the recall set includes at least two target objects;

the sequencing module is used for sequencing the at least two target objects to generate a sequencing result;

and the output module is used for outputting the at least two target objects according to the sequencing result.

In a third aspect, an embodiment of the present application provides a computer device, which includes a memory, a processor, and a computer program stored on the memory and executable on the processor, where the processor implements the steps in the object recommendation method when executing the computer program.

In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, on which a computer program is stored, where the computer program, when executed by a processor, implements the steps in the object recommendation method.

In the embodiment of the application, an object recommendation method is provided, which is used for cleaning order data and acquiring an object set corresponding to the order data; the set of objects includes at least two objects; determining the basic heat of each object in the object set; determining a heat characteristic of each object based on the basic heat of each object; ranking the heat characteristics of all the objects in the object set, and determining the objects corresponding to the heat characteristics in a preset ranking range as a recall set; the recall set includes at least two target objects; sequencing the at least two target objects to generate a sequencing result; according to the sorting result, the at least two target objects are output, and objects with heat can be obtained and recommended, so that the objects are recommended more accurately; and under the scene that the object is a commodity, the method can bring actual benefits to the advertisement media, and improves the user experience.

Drawings

Fig. 1 is a schematic network architecture diagram of an object recommendation system according to an embodiment of the present application;

fig. 2 is a first schematic flow chart illustrating an implementation of an object recommendation method according to an embodiment of the present application;

fig. 3A is a schematic flow chart illustrating an implementation process of an object recommendation method according to an embodiment of the present application;

fig. 3B is a schematic flow chart illustrating an implementation of the object recommendation method according to the embodiment of the present application;

fig. 4 is a schematic structural diagram of a component of an object recommendation device according to an embodiment of the present application;

fig. 5 is a hardware entity diagram of a computer device according to an embodiment of the present application.

Detailed Description

In order to make the objects, technical solutions and advantages of the embodiments of the present application clearer, the following will describe the specific technical solutions of the present application in further detail with reference to the accompanying drawings in the embodiments of the present application. The following examples are intended to illustrate the present application but are not intended to limit the scope of the present application.

Before further detailed description of the embodiments of the present application, terms and expressions referred to in the embodiments of the present application are described for better understanding of the present application, and the terms and expressions referred to in the embodiments of the present application are applied to the following explanations.

1) The hot degree characteristic refers to a hot sales degree of an object, and here, the object may be an advertisement, an application program, and the like.

2) A recall for a set of recommended objects.

3) The knowledge graph characteristic refers to mining the hidden knowledge entity of the object to obtain the internal knowledge relation of the object.

4) The user static characteristics refer to static characteristics such as gender, age, education level, purchasing power level, membership grade and the like of the user, and do not change with the change of the behavior of the user.

5) The user behavior characteristics refer to characteristics related to the user behavior, such as objects, categories and the like browsed, clicked and purchased by the user, and change along with the change of the user behavior.

The embodiment of the application can provide an object recommendation method, an object recommendation device, object recommendation equipment and a storage medium. In practical applications, the object recommendation method may be implemented by an object recommendation apparatus, and each functional entity in the object recommendation apparatus may be cooperatively implemented by hardware resources of a computer device (e.g., a terminal device, a server), such as computing resources like a processor, and communication resources (e.g., for supporting communications in various manners like optical cables and cellular).

The object recommendation method of the embodiment of the application can be applied to the object recommendation system shown in fig. 1, and as shown in fig. 1, the object recommendation system includes: a server 11 and a client 12.

Here, the server 11 cleans order data, and obtains an object set corresponding to the order data; the set of objects includes at least two objects; determining the basic heat of each object in the object set; determining a heat characteristic of each object based on the basic heat of each object; ranking the heat characteristics of all the objects in the object set, and determining the objects corresponding to the heat characteristics in a preset ranking range as a recall set; the recall set includes at least two target objects; sequencing the at least two target objects to generate a sequencing result; and outputting the at least two target objects according to the sequencing result. After the server 11 determines at least two target objects to be output, the target objects to be output are sent to the client 12 through the network; the client 12 displays at least two target objects; the user may operate on at least two target objects displayed by the client 12, such as: click, browse, purchase, etc.

Embodiments of an object recommendation method, an apparatus, a device and a storage medium according to the embodiments of the present application are described below with reference to a schematic diagram of an object recommendation system shown in fig. 1.

The embodiment of the application provides an object recommendation method, which is applied to a server, wherein the server can be a computer. The functions implemented by the method may be implemented by calling program code by a processor in a computer device, which may, of course, be stored in a computer storage medium, which may comprise at least a processor and a storage medium.

Fig. 2 is a schematic flow chart of an implementation of an object recommendation method according to an embodiment of the present application, and as shown in fig. 2, the method may include the following steps:

s201, cleaning order data to obtain an object set corresponding to the order data;

wherein the set of objects includes at least two objects.

The order data can have dirty data, the dirty data in the order data are cleaned, and an object set corresponding to the order data is obtained after the dirty data in the order data are cleaned;

here, the dirty data in the order data includes: dirty data of the bill swiping action, and dirty data corresponding to special commodities such as gifts, insurance and the like.

S202, determining the basic heat of each object in the object set;

here, after the object set corresponding to the order data is acquired, the basic heat of each object in the object set is determined.

Here, when determining the basic heat of each object in the object set, the basic heat may be determined according to order information and a time attenuation factor of the object. Wherein, the order information includes: the object in the order corresponds to the information of order quantity, total amount of transaction, commission, order time and the like.

Here, the degree of heat at the current time of the object in the order whose time is closer to the current time is larger based on the previous order data. According to the method and the device, the order quantity, the total volume of the deal and the commission corresponding to the object are linearly weighted, and the time attenuation factor is used for time attenuation of the heat according to the order time to obtain the basic heat of the object; wherein the basic heat can be obtained by formula (1):

in the formula (1), HotScoreBase_iRepresenting a base heat of the object i; s_iRepresenting a set of orders of the object within a set time period; beta is a time attenuation factor, the larger beta is, the more the basic heat scoring of the object pays more attention to the heat of the object in a short time, and if the influence of a relatively long-time object order on the basic heat scoring of the object needs to be considered, the beta can be properly reduced; t is_nowAnd T_order,jRespectively representing the current time and the time of the order j; OrderNum_j、GVM_j、CosFee_jRespectively representing the order quantity, GMV and commission of the object i in the order j; omega₁、ω₂、ω₃Linear weights representing the subject order volume, GMV and commission, respectively.

S203, determining the heat degree characteristic of each object based on the basic heat degree of each object;

here, a heat degree feature of each object is determined based on the base heat degree of each object, and the heat degree feature characterizes a degree of hot sales of the object. The object can be advertisement content such as goods and application programs.

Here, the heat characteristic of each object in the object set may determine the heat characteristic of the object according to the base heat of each object and the rank of the base heat in the base heat set of the target category.

S204, ranking the heat characteristics of all the objects in the object set, and determining the objects corresponding to the heat characteristics in a preset ranking range as a recall set;

here, the recall set includes at least two target objects. After the heat degree characteristics of all the objects in the object set are obtained, the heat degree characteristics of all the objects are sorted, and the objects with the heat degree characteristics in a preset sorting range are used as target objects and are placed into a recall set.

Here, the preset sorting range may be set according to practical application situations, for example: the preset sequencing range may be: the first two or the first three are sorted, and the preset sorting range is not limited in any way in the embodiment of the application.

Such as: the object set comprises: object 1, object 2, object 3, object 4, object 1 having a heat signature of H1, object 2 having a heat signature of H2, object 3 having a heat signature of H3, object 4 having a heat signature of H4, wherein H2> H1> H4> H3, the set range is: the first two, therefore, object 2, object 1 are determined to be a recall.

For another example: the object set comprises: object 1, object 2, object 3, object 4, object 1 having a heat signature of H1, object 2 having a heat signature of H2, object 3 having a heat signature of H3, object 4 having a heat signature of H4, wherein H1> H2> H3> H4, the set range is: the first three, therefore, object 1, object 2, object 3 are determined to be a recall.

The setting range may be set according to actual conditions, and the setting range is not limited in any way in the embodiments of the present application.

S205, sequencing the at least two target objects to generate a sequencing result;

after the recall set is determined, at least two target objects in the recall set are ranked to obtain a ranking result; when at least two target objects in the recall set are sorted, the target objects may be sorted in a forward order or in a reverse order, and the sort type is not limited in any way in the embodiment of the present application.

S206, outputting the at least two target objects according to the sorting result.

Here, after the sorting result is obtained, the target object is output according to the sorting result. When the target objects are output according to the sorting result, all the target objects in the recall set may be output, or part of the target objects in the recall set may be output.

Such as: the recall set includes: target object 1, target object 2, target object 3, target object 4, target object 5; ordering the objects results in: target object 2, target object 3, target object 1, target object 5, and target object 4, and all target objects are output according to the sorting result, that is, target object 2, target object 3, target object 1, target object 5, and target object 4 are output.

For another example: the recall set includes: target object 1, target object 2, target object 3, target object 4, target object 5; ordering the objects results in: target object 1, target object 2, target object 3, target object 4, and target object 5, and the top three ranked target objects are output, that is, output target object 1, target object 2, and target object 3.

In one example, sorting at least two target objects in the recall set, and outputting the at least two target objects according to a sorting result of the sorting comprises: sequencing at least two target objects in the recall set to obtain a first sequencing result; adjusting the first sorting result based on the category to which the target object corresponding to the first sorting result belongs to obtain a second sorting result; and outputting the target object corresponding to the second sequencing result in the set range.

Firstly, at least two target objects in a recall set are sorted to obtain a first sorting result; determining the category to which the target object in the first sequencing result belongs, scattering the target object belonging to the same category in the first sequencing result, adjusting to obtain a second sequencing result, and outputting the target object in the set range.

Such as: the category to which the target object 1 belongs is clothing; the category to which the target object 2 belongs is clothing; the category to which the target object 3 belongs is shoes, and the category to which the target object 4 belongs is catering; sequencing the target objects to obtain a first sequencing result: the target object 1, the target object 2, the target object 3, and the target object 4 are further classified according to the categories to which the target object 1, the target object 2, the target object 3, and the target object 4 belong, wherein the target object 1 and the target object 2 belong to the same category, and the target object 3, the target object 1, and the target object 2 belong to different categories, so that the target object 2 and the target object 3 are exchanged to scatter the categories to which the target object 1, the target object 2, the target object 3, and the target object 4 belong, and a second ranking result is obtained: target object 1, target object 3, target object 2, and target object 4, and the target objects ranked in the top two are output, that is, target object 1 and target object 3 are output.

In the embodiment of the application, order data are cleaned, and an object set corresponding to the order data is obtained; the set of objects includes at least two objects; determining the basic heat of each object in the object set; determining a heat characteristic of each object based on the basic heat of each object; ranking the heat characteristics of all the objects in the object set, and determining the objects corresponding to the heat characteristics in a preset ranking range as a recall set; the recall set includes at least two target objects; sequencing the at least two target objects to generate a sequencing result; according to the sorting result, the at least two target objects are output, and the objects with heat can be obtained and recommended, so that the objects are recommended more accurately; and under the scene that the object is a commodity, the method can bring actual benefits to the advertisement media, and improves the user experience.

In one embodiment, determining the heat degree characteristic of each object based on the basic heat degree of each object comprises: for each object in the set of objects, performing the following: determining the heat characteristic of the object according to the basic heat and the ranking of the basic heat in a basic heat set of a target category; the target category is a category to which the object belongs; the basic heat set is the combination of the basic heat of each object under the target category.

Here, when the heat characteristics of each object in the object set are acquired, the basic heat of each object in the object set is first calculated, and then the heat characteristics of the objects are determined according to the basic heat and the ranking of the basic heat in the basic heat set of the target category.

Here, the object order situations between different categories may be relatively different, and it is not significant to directly compare the basic heat of different categories, so it is necessary to compare the heat characteristics of objects of the same category. When determining the heat degree characteristics of objects of the same class, firstly determining the heat degree ranking of the basic heat degree in the combination of the basic heat degrees of all the objects under the class to which the objects belong; and determining the heat characteristic of the object based on the basic heat and the heat ranking of the object.

Here, in determining the heat characteristic of the object, equation (2) may be employed:

in the formula (2), HotScore_iRepresenting the heat characteristic, Rank, of the object i_iIndicating the basic heat of an object i, HotScoreBase_iRanking under their category.

In the embodiment of the application, according to the basic heat and the ranking of the basic heat in the basic heat set of the target category, the heat characteristic of the object is determined, so that the real-time hot sales degree of the object is determined.

In an embodiment, the ranking of at least two target objects in the recall set includes: inputting the input characteristics of each target object in the at least two target objects into a recommendation model to obtain a recommendation value corresponding to each target object; the input features include at least one of: object characteristics, knowledge graph characteristics, user static characteristics and user behavior characteristics; and sequencing the at least two target objects according to the recommended value.

Here, when at least two target objects in the recall set are ranked, the input features of each of the at least two target objects in the recall set are input into the recommendation model, and a recommendation value corresponding to each target object is obtained. And sequencing at least two target objects in the recalls according to the recommended values.

Here, the input features include: object characteristics, knowledge graph characteristics, user static characteristics and user behavior characteristics; wherein the object features include: features such as heat features, price of the object, brand of the object, discount of the object, etc.; the knowledge graph features represent objects related to the target object or categories to which the objects belong; the user static features include: the user's own static attributes such as the user's gender, age, education level, purchasing power level, membership grade, sales promotion sensitivity, store of interest, etc.; the user behavior characteristics comprise: the category, object and other dynamic attributes of the user browsing, clicking, buying, purchasing and other activities are related to the user activities.

Here, when the input feature of each of the at least two target objects is input into the recommendation model, the input feature may be one of an object feature, a knowledge graph feature, a user static feature, and a user behavior feature, or may be a combination of any of the object feature, the knowledge graph feature, the user static feature, and the user behavior feature.

Such as: the recommended model is as follows: and f (x) ═ ω x + b, where x is the object feature and the feature of the target object 1, the object feature and the feature of the target object 2, and the object feature and the feature of the target object 3, and is input to the recommendation model, recommendation values f corresponding to the target object 1, the target object 2, and the target object 3 are obtained, respectively₁(x)、f₂(x)、f₃(x)。

For another example: the recommended model is as follows: (x) ω x + b, and the object characteristics of the target object 1, the object characteristics of the target object 2, the object characteristics of the target object 3, and the object characteristics of the target object 4 are input as x into the recommendation model to obtain the target object 1, the target object 2, and the target object 4, respectivelyRecommended value f corresponding to target object 3 and target object 4₁(x)、f₂(x)、f₃(x) And f₄(x)。

Also for example: the recommended model is as follows: and f (x) ═ ω x + b, inputting the object features, the knowledge graph features, the user static features and the user behavior features of the target object 1 and the object features, the knowledge graph features, the user static features and the user behavior features of the target object 2 as x into a recommendation model, and obtaining recommendation values f corresponding to the target object 1 and the target object 2 respectively₁(x) And f₂(x)。

Here, before the input feature of each of at least two target objects in the recall set is input into the recommendation model and a recommendation value corresponding to the target object is obtained, the method further includes: and training the recommendation model according to the input characteristics of each target object in the recall set, and updating the parameters of the recommendation model.

Here, after obtaining the recommendation value corresponding to each of the at least two target objects in the recall set, the at least two target objects are sorted based on the recommendation value.

Such as: the recommended value corresponding to the target object 1 is f₁(x) The recommended value corresponding to the target object 2 is f₂(x) The recommended value corresponding to the target object 3 is f₃(x) The recommended value corresponding to the target object 4 is f₄(x) For the recommended value f₁(x)、f₂(x)、f₃(x) And f₄(x) Sorting to obtain a sorting result: f. of₃(x)、f₂(x)、f₁(x)、f₄(x)。

In the embodiment of the application, the recommendation values corresponding to the target objects are obtained based on the input features and the recommendation model of each of the at least two target objects in the recall set, and the at least two target objects are sorted based on the recommendation values, so that the target objects which are hot sold in a specific range can be recommended and output, and the user experience is improved.

In an embodiment, the knowledge-graph features include a first collocation and/or a second collocation, the method further comprising: determining a first collocation relationship between the target object and a first associated object based on behavior data of a user; and/or determining a second collocation relationship between the target object and a second associated object according to the semantics of the object description of the target object.

Here, when determining the feature of the target object, the target object itself or the historical behavior of the user corresponding to the target object may be used as a seed set, and the seed set is propagated and diffused outward to obtain the object having a relationship with the target object or the category to which the object belongs.

Here, determining the knowledge-graph characteristics of the target object may include at least one of:

firstly, determining a first collocation relationship between a target object and a first associated object based on behavior data of a user.

Here, the collocation relationship between the target object and the first associated object is determined based on behavior data of a user such as a search behavior, a click behavior, a purchase behavior, a browse behavior, and the like. Such as: the target object is a skirt, the user clicks the shoes, and the server determines that the first associated object in matching relationship with the skirt is the shoes.

In an example, determining a first collocation relationship between the target object and the first associated object based on the behavioral data of the user includes: and determining the collocation relationship between the category to which the target object belongs and the category to which the first associated object belongs based on the behavior data of the user.

Here, the collocation relationship between the category to which the target object belongs and the category to which the first related object belongs is determined based on behavior data of the user such as a search behavior, a click behavior, a purchase behavior, a browse behavior, and the like. Such as: the target object is a skirt, the category to which the skirt belongs is clothes, the user clicks the necklace, the category to which the necklace belongs is the clothes, and the server determines the category associated with the clothes to be the clothes.

And secondly, determining a second collocation relationship between the target object and a second associated object according to the semantics of the object description of the target object.

Here, the semantics of the object description of the target object include: the semantics of the object such as model, color collocation, taste, specification, style and the like determine the collocation relationship between the target object and the second associated object according to the semantics of the object description of the target object.

Such as: the model of the object: if the user has recently purchased iphone10, the determined associated object is the phone film of iphone10, rather than the phone film of iphone 6; for another example: style matching of objects: if the user has recently purchased a skirt, the associated object is determined to be a hat or shoes or the like suitable for the skirt.

In the embodiment of the application, the knowledge graph characteristics corresponding to the target object can be determined according to the behavior data of the user or the semantics of the object description of the target object, and the knowledge graph characteristics are used as a basis for calculating the recommendation value corresponding to the target object.

In an embodiment, the inputting the input feature of each of the at least two target objects into the recommendation model to obtain the recommendation value corresponding to each of the target objects includes: inputting the input characteristics of each target object in the at least two target objects into a gradient lifting decision tree (GBDT) model to obtain the output characteristics of the GBDT model; the input features include at least one of: the method comprises the following steps of (1) obtaining object characteristics, a first matching relation, a second matching relation, user static characteristics and user behavior characteristics; and inputting the output characteristics of the GBDT model into a logistic regression LR model to obtain a recommended value corresponding to each target object.

Here, when calculating the recommended value of each of the at least two target objects, the input feature of each target object may be input to the GBDT model to obtain the output feature of the GBDT model, and the output feature of the GBDT model may be used as the input feature of the LR model to obtain the recommended value corresponding to each target object.

Here, the input features include: when the input features of each target object are input into the GBDT model, any one or any combination of the object features, the first collocation relationship, the second collocation relationship, the user static features and the user behavior features can be input into the recommendation model.

Such as: GBDT model is f (x) ax + c, LR model is f (y) ey + f, object features and knowledge graph features of target object 1 and object features and knowledge graph features of target object 2 are x, GBDT model f (x) ax + c is input to obtain output features f (x) of GBDT model, f (x) is input y of LR model f (y) ey + f, and recommended values f (x) corresponding to target object 1 and target object 2 are obtained₁(y)、f₂(y)。

Here, before obtaining the recommended value corresponding to the target object using the GBDT model and the LR model, the GBDT model and the LR model need to be trained, and parameters of the GBDT model and the LR model need to be updated.

Here, user static features, such as: gender, age, education level, etc. of the user, subject characteristics such as price, discount, popularity characteristics, etc. of the subject, user dynamic characteristics, such as: and browsing, clicking, buying, paying attention, the category and the object of the purchased object, and the knowledge graph characteristics are used as input characteristics of the GBDT model, whether the user clicks the target object is used as output characteristics of the GBDT model, the GBDT model is trained, and the parameters of the GBDT model are updated. And then, training the LR model by taking the output characteristic of the GBDT model as the input characteristic of the LR model and taking the recommended value corresponding to the target object as the output characteristic of the LR model, and updating the parameters of the LR model.

In the embodiment of the application, the recommendation value corresponding to the target object can be obtained according to the GBDT model and the LR model, the target object is output based on the recommendation value, the object can be recommended more accurately, and user experience is improved.

In one embodiment, the cleaning the order data includes: identifying the swiped line as a corresponding invalid order; and screening out the invalid objects corresponding to the invalid orders.

Here, the dirty data may be mixed in the object order due to commercial activities such as e-commerce billing, gifts, advertisement cheating and the like, so the dirty data in the object order is cleaned before the popularity characteristics of each object in the object set are acquired; wherein the cleaning of the dirty data comprises: and cleaning the corresponding invalid order of the brush line.

Here, when an invalid order is purged, the order of the order-flushing behavior is recognized as an invalid order, and an invalid object corresponding to the invalid order is screened from the object set.

Here, the machine learning algorithm obtained by training identifies the corresponding invalid object as a brush line. Wherein the machine learning algorithm may include: an isolated forest isolated point detection algorithm, a supervised learning algorithm and the like.

The method comprises the steps of identifying invalid objects by using an isolated forest isolated point detection algorithm, constructing an isolated tree according to attributes of order quantity, total volume of deals, commission, order placement time distribution, order placement address distribution and the like of the objects, calculating the path length of each path of the isolated tree, determining an abnormal value according to the path length, and determining the objects as invalid objects when the abnormal value is within a set range.

It should be noted that, the isolated forest isolated point detection algorithm is an example, and in the embodiment of the present application, a plurality of machine learning algorithms may be used to identify an invalid object, and the embodiment of the present application does not limit the machine learning algorithms at all.

In the embodiment of the application, before the heat characteristics of each object in the object set are obtained, the invalid objects corresponding to the invalid orders can be screened out from the object set, and invalid data are prevented from being mixed in the object orders.

In one embodiment, the cleaning the order data includes: analyzing order data of a user; determining a designated object in order placing objects corresponding to the order data, wherein the designated object is an object without a deal amount; and screening out the specified object.

Here, the flushing dirty data further includes: dirty data of a designated object such as a gift is cleaned. And the server analyzes the order data of the user, determines the object as a specified object when the object in the order is the object without the transaction amount, and screens out the specified object from the object set.

Here, when the object is determined to be a designated object, the object may be obtained through a word segmenter and a word2vec algorithm model. Firstly, segmenting the object name according to a word segmenter to form a word vector of the object name, then converting the word vector of the object name into a numerical vector according to a word2vec algorithm, and calculating the corresponding vector distance to determine the specified object.

Such as: the object name is: the method comprises the steps of classifying words of a Jageyi Korean gold snail stock solution gift by an ik word classifier to form word vectors, obtaining word vectors of the Jageyi Korean, Korea, gold, snail, stock solution, skin care products and gift, obtaining a numerical value vector of an object by adopting a word2vec algorithm according to the word vectors of a whole object library, calculating a vector distance corresponding to the numerical value vector, and identifying the corresponding object as a designated object when the vector distance is smaller than a certain threshold value.

In the embodiment of the application, the designated objects without the deal amount can be screened from the object set before the heat characteristics of all the objects in the object set are acquired, and invalid data is prevented from being mixed in the object order.

Hereinafter, the object recommendation method provided in the embodiment of the present application is further described by taking product recommendation of CPS advertisement as an example.

In the e-commerce advertisement type, the existing advertisement formats mainly include Cost Per Time period (CPT) advertisement, presentation Cost (CPM) advertisement, pay Per Click (CPC) advertisement, CPS advertisement, and the like according to different charging manners. The CPS advertisement has the advantages of large long tail flow base number and low flow cost. CPS advertisements are generally applied to scenes with large long tail traffic and low traffic quality.

The CPS advertisement application scenes mainly include scenes with large long tail flow, such as a new advertisement page, an express cabinet and the like during code scanning payment, and the CPS advertisement has the advantage of large flow base number and simultaneously faces the challenges of relatively low final conversion and difficulty in bringing actual benefits to advertisement media. In a commodity recommendation scene of the CPS advertisement, the final order conversion can bring actual benefits to media, and the CPS advertisement is charged according to the order amount, so the conversion effect of the advertisement is more emphasized in the advertisement form.

Due to the fact that the CPS advertisement carries out charging according to order conversion, under the CPS advertisement scene, the real-time hot sales degree of the commodity is emphasized, the hot sales degree of the commodity has high timeliness, and the old hot sales commodity can be replaced by the new hot sales commodity within short time. In this case, it is particularly important to mine the popularity of hot-sold products in real time or near real time, but the popularity of such products is far from being mined in the existing product recommendations for CPS advertisements.

Under the modern electronic market scene, commodities contain rich knowledge entities, and users tend to click and purchase commodities which are closely related to the commodities and have related knowledge with previous behaviors, such as browsing, clicking, buying, adding and purchasing. In the conventional association model, only surface association between commodities is generally considered, and the mining of related knowledge entities and relationships of the commodities through a knowledge graph can mine more intrinsic and hidden knowledge associations of the commodities and provide more accurate commodity recommendation.

Therefore, in the product recommendation of the CPS advertisement, in addition to consideration of the click intention of the user and the collocation relationship between the products, attention needs to be paid to the heat mining of the recommended products themselves.

At present, methods such as collaborative filtering, GBDT and knowledge maps are widely applied to commodity recommendation, but a set of comparison system and a useful recommendation method do not exist for commodity recommendation of CPS advertisements, so that the embodiment of the application provides a knowledge map-based personalized recommendation method for CPS hot-sell commodities to perform more accurate CPS commodity recommendation for media.

The embodiment of the application provides a knowledge graph-based personalized recommendation method for CPS hot-sell commodities based on specific scenes and requirements of commodity recommendation of CPS advertisements. Firstly, digging out current hot sold commodities as a recall set recommended by the CPS commodities based on commodity popularity mining, and ensuring the quality and popularity of the recommended commodities; secondly, establishing commodity knowledge entities based on a knowledge graph method, mining the intrinsic hidden knowledge relation among commodities, and constructing a richer and more accurate commodity recommendation network; and finally, mining the personalized preference of the user by adopting a GBDT + LR method based on the user behavior data, and carrying out personalized recommendation on the user.

The method of the embodiment of the application comprises the following steps:

the method comprises the following steps: commodity order data cleaning

In the actual scene of the CPS advertisement, the commercial activities such as e-commerce billing, gifts, advertisement cheating and the like can cause dirty data to be mixed in the commodity order data, the dirty data needs to be cleaned firstly, and the cleaning of the data mainly has the following two aspects:

A. the brush single row is: according to attributes of order quantity, total transaction Volume (GMV), commission, order placing time distribution, order placing address distribution and the like of the commodity, a regression and other machine learning algorithm framework is constructed to train the commodity, identify the order-brushing commodity and filter the commodity.

B. Special commodities such as gifts: gifts, insurance, and the like belong to special commodities which have high order quantity but are not suitable for a recommended scene, and the special commodities should be filtered. According to the commodity names, such special commodities or similar commodities are mined out through an ik word segmentation device and a word2vec algorithm model and are filtered.

Step two: hot-sell commodity digging

In the CPS scenario, the commodity popularity of the commodities in the recall set has a large influence on the final conversion of the recommendation result, however, due to the influence of factors such as killing by seconds, live broadcast, coupon timeliness, and the like, the popularity of the commodities has short timeliness, and the commodities with high popularity in the past several hours are highly likely to lose popularity at present. Therefore, in the embodiment of the present application, a comprehensive hot-market commodity mining strategy is proposed:

A. the commodity basic heat grading strategy considering time attenuation based on the commodity order comprises the following steps:

in one aspect, the order status of the goods, such as: the order quantity, the GMV and the commission represent the heat of the commodity, on the other hand, the order with the time closer to the current time has the heat of the commodity at the current time, so based on the order data of the past 12 hours, the embodiment of the application linearly weights the order quantity, the GMV and the commission of the commodity and performs time attenuation according to the order time to obtain the basic heat scoring strategy of the commodity, as shown in formula (1):

in the formula (1), HotScoreBase_iRepresents the base heat score of the commodity i; s_iA set of orders representing the items over the past 12 hours; beta is a time attenuation factor, the larger beta is, the more important the basic heat scoring of the commodities is, if the influence of the commodity orders for a relatively long time on the basic heat scoring of the commodities needs to be considered, the beta can be properly reduced; t is_nowAnd T_order,jRespectively representing the current time and the time of the order j; OrderNum_j、GVM_j、CosFee_jRespectively representing the order quantity, GMV and commission of the commodity i in the order j; omega₁、ω₂、ω₃Linear weights representing the volume of orders for goods, GMV and commission, respectively.

B. The commodity category scattering strategy based on category ranking comprises the following steps:

the commodity order conditions between different categories are probably different than great, and the comparison significance is not great by directly scoring the basic heat of different categories, so that a commodity category scattering strategy based on category ranking is designed in the embodiment of the application, as shown in formula (2):

in the formula (2), HotScore_iRepresents the final heat score, Rank, of the commodity i_iBasic score HotScoreBase representing item i_iRanking under its third category, HotScoreBase_iIs the commodity heat base score obtained by the formula (1).

C. According to the formula (1) and the formula (2), top N hot commodities which are scattered and obtained and take time attenuation into consideration are mined and used as a CPS hot commodity recall candidate set, and the quality and the hot degree of the commodities in the recall candidate set are guaranteed.

Step three: commodity knowledge linkage mining

In the modern E-commerce big data background, the internal knowledge relation between commodities can be mined through a knowledge graph. According to the method, the historical behaviors of each user on the E-commerce commodity are used as seed sets on the knowledge graph by adopting an outward propagation method, and the outward propagation is expanded along the knowledge graph. The mining of the commodity knowledge relation in the embodiment of the application mainly comprises the following two aspects:

A. and (3) commodity knowledge mining based on user behaviors: mining the related collocation relationship between commodities and between categories according to user behaviors such as user common purchasing behavior, search common clicking behavior, similar browsing behavior and the like;

B. and (3) mining commodity knowledge based on the self semantics of the commodity: and mining the self semantic collocation relationship of the commodities according to commodity semantics such as the model, color collocation, taste, specification, style and the like of the commodities.

And mining the intrinsic knowledge relation of the commodities according to the two points to obtain the intrinsic knowledge relation of the commodities as the input characteristics of the CPS commodity personalized recommendation algorithm.

Step four: personalized merchandise recommendation

The embodiment of the application adopts a GBDT + LR algorithm framework to carry out personalized recommendation on the commodities.

Users often have a significant tendency to click or purchase goods; in a CPS advertisement scene, the tendency of a user is mainly related to four factors, namely, the user's own attributes, the user's actions such as browsing, clicking, buying, purchasing and the like, the product's own attributes, and the knowledge link attributes of the product, so the embodiment of the present application performs algorithm model training by using the following four attributes as input features of the GBDT + LR algorithm framework:

A. the user's own static attributes such as the user's gender, age, education level, purchasing power level, membership grade, sales promotion sensitivity, store of interest, etc.;

B. the user's own dynamic preference attributes such as categories, brands and the like of recent browsing, clicking, buying, purchasing and the like are closely related to time, and thus, the user's own dynamic preference attributes can be further classified into the user's own dynamic preference attributes such as the user's own preferences such as recent 1 day browsing, clicking, buying, purchasing brand preference, recent 1 day browsing, clicking, buying, purchasing category preference, recent 3 days browsing, clicking, purchasing category preference and the like;

C. commodity attributes such as price, discount, popularity, brand, category, etc. of the commodity;

D. and related commodity knowledge link attributes of the current commodity, related three-level categories, model collocation and the like are obtained by mining the commodity knowledge link attributes in the third step.

And training the GBDT + LR algorithm according to the attribute characteristics, and updating the parameters of the GBDT + LR algorithm. After the parameters of the GBDT + LR algorithm are obtained, commodity attributes such as user static attributes, user dynamic behavior attributes and heat characteristics and commodity intrinsic knowledge are input into the GBDT + LR algorithm in a linkage mode, and a sexualization sequencing result is obtained. When obtaining the personalized ranking result, as shown in fig. 3A, the commodity attributes 33 such as the user static attribute 31, the user dynamic behavior attribute 32, the heat characteristic, and the like, and the commodity intrinsic knowledge link 34 are input into the GBDT + LR algorithm 35 to obtain the personalized ranking result 36. Among them, the user static attribute 31 may include: information such as gender, age, etc.; the user dynamic behavior attributes 32 may include: browsing information, click information, purchase information, etc.; the commodity attributes 33 include: price, discount, heat characteristics, etc.

As shown in fig. 3B, the recommendation method of the embodiment of the present application includes the following steps:

step 301: cleaning the commodity order data;

step 302: carrying out basic heat time attenuation scoring on the commodities in the cleaned commodity order;

step 303: obtaining heat characteristics based on the basic heat and the category ranking;

step 304: taking the commodities with the heat characteristics within a set range as a recommended recall candidate set;

here, the recall candidate set is recommended as the recall set in the above-described embodiment.

Step 305: acquiring commodity attributes, user static attributes, user dynamic attributes and intrinsic knowledge link attributes;

wherein the commodity attributes include: price, discount, heat characteristics, etc.

Step 306: inputting the user static attribute, the user dynamic attribute, the commodity attribute and the intrinsic knowledge link attribute into a GBDT + LR algorithm;

step 307: and obtaining a personalized sequencing result.

The CPS hot-selling commodity recommendation method based on the knowledge graph is applied to CPS alliance two-in-one scene commodity recommendation, and finally, the advertisement income (effective cost per mile, eCPM) and GMV (GMV) or the visit volume (Page View, PV) obtained by displaying each thousand times of the scene are respectively increased by 75% and 50%.

By the recommendation method provided by the embodiment of the application, the following technical effects can be achieved:

1) the method has the advantages that the current commodities with higher popularity can be mined out to serve as the recalling set recommended by the CPS commodities, the popularity of the commodities in the recalling set is guaranteed, and therefore the actual conversion of CPS commodity recommendation is promoted.

2) And mining the knowledge entities hidden in the commodities based on a knowledge graph method to obtain the intrinsic knowledge relation of the commodities, thereby carrying out richer commodity recommendation.

3) Based on 1) and 2), performing personalized commodity recommendation on the user by adopting a GBDT + LR algorithm according to the user behavior data.

Based on the foregoing embodiments, the present application provides an object recommendation apparatus, where the apparatus includes modules and units included in the modules, and may be implemented by a processor in a computer device; of course, the implementation can also be realized through a specific logic circuit; in implementation, the processor may be a Central Processing Unit (CPU), a Microprocessor (MPU), a Digital Signal Processor (DSP), a Field Programmable Gate Array (FPGA), or the like.

Fig. 4 is a schematic structural diagram of an object recommendation device according to an embodiment of the present application, and as shown in fig. 4, the device 40 includes: an acquisition module 401, a first determination module 402, a second determination module 403, a third determination module 404, a sorting module 405, and an output module 406; wherein the content of the first and second substances,

an obtaining module 401, configured to clean order data and obtain an object set corresponding to the order data; the set of objects includes at least two objects;

a first determining module 402, configured to determine a base heat of each object in the object set;

a second determining module 403, configured to determine a heat degree characteristic of each object based on the basic heat degree of each object;

a third determining module 404, configured to rank the heat characteristics of each object in the object set, and determine an object corresponding to the heat characteristic within a preset ranking range as a recall set; the recall set includes at least two target objects;

a sorting module 405, configured to sort the at least two target objects, and generate a sorting result;

an output module 406, configured to output the at least two target objects according to the sorting result.

In an embodiment, the second determining module 403 is configured to determine the heat characteristic of the object according to the base heat and the rank of the base heat in the base heat set of the target category; the target category is a category to which the object belongs; the basic heat set is the combination of the basic heat of each object under the target category.

In one embodiment, the sorting module 405 includes: a first determining unit and a sorting unit; wherein the content of the first and second substances,

the first determining unit is configured to input the input feature of each of the at least two target objects into a recommendation model to obtain a recommendation value corresponding to each of the target objects; the input features include at least one of: object characteristics, knowledge graph characteristics, user static characteristics and user behavior characteristics;

the sorting unit is used for sorting the at least two target objects according to the recommended value.

In one embodiment, the apparatus 40 further comprises: a fourth determination module and a fifth determination module; wherein the content of the first and second substances,

the fourth determining module is configured to determine a first collocation relationship between the target object and the first associated object based on behavior data of a user;

the fifth determining module is configured to determine a second collocation relationship between the target object and a second associated object according to the semantic meaning of the object description of the target object.

In an embodiment, the first determination unit includes: a first determining subunit and a second determining subunit; wherein the content of the first and second substances,

the first determining subunit is configured to input the input feature of each of the at least two target objects into a gradient boosting decision tree GBDT model to obtain an output feature of the GBDT model; the input features include at least one of: the method comprises the following steps of (1) obtaining object characteristics, a first matching relation, a second matching relation, user static characteristics and user behavior characteristics;

and the second determining subunit is used for inputting the output characteristics of the GBDT model into a Logistic Regression (LR) model to obtain a recommended value corresponding to each target object.

In one embodiment, the obtaining module 401 includes: the device comprises an identification unit and a first screening unit; wherein the content of the first and second substances,

the identification unit is used for identifying the corresponding invalid order as the brushing line;

the first screening unit is configured to screen an invalid object corresponding to the invalid order from the object set.

In an embodiment, the obtaining module 401 further includes: the analysis unit, the second determination unit and the second screening unit; wherein the content of the first and second substances,

the analysis unit is used for analyzing the order data of the user;

the second determining unit is configured to determine a designated object in order placing objects corresponding to the order data, where the designated object is an object without a deal amount;

the second screening unit is used for screening the specified object from the object set.

The above description of the apparatus embodiments, similar to the above description of the method embodiments, has similar beneficial effects as the method embodiments. For technical details not disclosed in the embodiments of the apparatus of the present application, reference is made to the description of the embodiments of the method of the present application for understanding.

It should be noted that, in the embodiment of the present application, if the object recommendation method is implemented in the form of a software functional module and sold or used as a standalone product, the object recommendation method may also be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the embodiments of the present application may be essentially implemented or portions thereof contributing to the related art may be embodied in the form of a software product stored in a storage medium, and including several instructions for enabling a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the methods described in the embodiments of the present application. And the aforementioned storage medium includes: various media capable of storing program codes, such as a usb disk, a removable hard disk, a Read Only Memory (ROM), a magnetic disk, or an optical disk. Thus, embodiments of the present application are not limited to any specific combination of hardware and software.

Correspondingly, the present application provides an apparatus, that is, a computer apparatus, including a memory and a processor, where the memory stores a computer program executable on the processor, and the processor executes the program to implement the steps in the object recommendation method provided in the foregoing embodiments.

Accordingly, embodiments of the present application provide a storage medium, that is, a computer-readable storage medium, on which a computer program is stored, and the computer program, when executed by a processor, implements the steps in the object recommendation method provided in the above embodiments.

Here, it should be noted that: the above description of the storage medium and device embodiments is similar to the description of the method embodiments above, with similar advantageous effects as the method embodiments. For technical details not disclosed in the embodiments of the storage medium and apparatus of the present application, reference is made to the description of the embodiments of the method of the present application for understanding.

It should be noted that fig. 5 is a schematic hardware entity diagram of a computer device according to an embodiment of the present application, and as shown in fig. 5, the computer device 500 includes: a processor 501, at least one communication bus 502, at least one external communication interface 504, and memory 505. Wherein the communication bus 502 is configured to enable connective communication between these components. In an embodiment, the computer device may further comprise: the user interface 503, the user interface 503 may include a display screen, and the external communication interface 504 may include a standard wired interface and a wireless interface.

The Memory 505 is configured to store instructions and applications executable by the processor 501, and may also buffer data (e.g., image data, audio data, voice communication data, and video communication data) to be processed or already processed by the processor 501 and modules in the computer device, and may be implemented by a FLASH Memory (FLASH) or a Random Access Memory (RAM).

It should be appreciated that reference throughout this specification to "one embodiment" or "an embodiment" means that a particular feature, structure or characteristic described in connection with the embodiment is included in at least one embodiment of the present application. Thus, the appearances of the phrases "in one embodiment" or "in an embodiment" in various places throughout this specification are not necessarily all referring to the same embodiment. Furthermore, the particular features, structures, or characteristics may be combined in any suitable manner in one or more embodiments. It should be understood that, in the various embodiments of the present application, the sequence numbers of the above-mentioned processes do not mean the execution sequence, and the execution sequence of each process should be determined by its function and inherent logic, and should not constitute any limitation to the implementation process of the embodiments of the present application. The above-mentioned serial numbers of the embodiments of the present application are merely for description and do not represent the merits of the embodiments.

It should be noted that, in this document, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other like elements in a process, method, article, or apparatus that comprises the element.

In the several embodiments provided in the present application, it should be understood that the disclosed apparatus and method may be implemented in other ways. The above-described embodiment of the apparatus is only illustrative, for example, the splitting of the unit is only a logical function splitting, and there may be other splitting manners in actual implementation, such as: multiple units or components may be combined, or may be integrated into another system, or some features may be omitted, or not implemented. In addition, the coupling, direct coupling or communication connection between the components shown or discussed may be through some interfaces, and the indirect coupling or communication connection between the devices or units may be electrical, mechanical or other forms.

The units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units; can be located in one place or distributed on a plurality of network units; some or all of the units can be selected according to actual needs to achieve the purpose of the solution of the embodiment.

In addition, all functional units in the embodiments of the present application may be integrated into one processing unit, or each unit may be separately regarded as one unit, or two or more units may be integrated into one unit; the integrated unit can be realized in a form of hardware, or in a form of hardware plus a software functional unit.

Those of ordinary skill in the art will understand that: all or part of the steps for realizing the method embodiments can be completed by hardware related to program instructions, the program can be stored in a computer readable storage medium, and the program executes the steps comprising the method embodiments when executed; and the aforementioned storage medium includes: various media that can store program codes, such as a removable Memory device, a Read Only Memory (ROM), a magnetic disk, or an optical disk.

Alternatively, the integrated units described above in the present application may be stored in a computer-readable storage medium if they are implemented in the form of software functional modules and sold or used as independent products. Based on such understanding, the technical solutions of the embodiments of the present application may be essentially implemented or portions thereof contributing to the related art may be embodied in the form of a software product stored in a storage medium, and including several instructions for enabling a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the methods described in the embodiments of the present application. And the aforementioned storage medium includes: a removable storage device, a ROM, a magnetic or optical disk, or other various media that can store program code.

The above description is only for the embodiments of the present application, but the scope of the present application is not limited thereto, and any person skilled in the art can easily conceive of changes or substitutions within the technical scope of the present application, and shall be covered by the scope of the present application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims.

Claims

1. An object recommendation method, characterized in that the method comprises:

determining the basic heat of each object in the object set;

sequencing the at least two target objects to generate a sequencing result;

2. The method of claim 1, wherein determining the heat profile of each of the objects based on the base heat of each of the objects comprises:

for each object in the set of objects, performing the following:

determining the heat characteristic of the object according to the basic heat and the ranking of the basic heat in a basic heat set of a target category; the target category is a category to which the object belongs; the basic heat set is the combination of the basic heat of each object under the target category.

3. The method of claim 1, wherein the ordering the at least two target objects comprises:

inputting the input characteristics of each target object in the at least two target objects into a recommendation model to obtain a recommendation value corresponding to each target object; the input features include at least one of: object characteristics, knowledge graph characteristics, user static characteristics and user behavior characteristics;

and sequencing the at least two target objects according to the recommended value.

4. The method of claim 3, wherein the knowledge-graph features comprise a first collocation and/or a second collocation, the method further comprising:

determining a first collocation relationship between the target object and a first associated object based on behavior data of a user; and/or the presence of a gas in the gas,

and determining a second collocation relationship between the target object and a second associated object according to the semantics of the object description of the target object.

5. The method of claim 4, wherein the inputting the input features of each of the at least two target objects into a recommendation model to obtain a recommendation value corresponding to each of the at least two target objects comprises:

inputting the input characteristics of each target object in the at least two target objects into a gradient lifting decision tree (GBDT) model to obtain the output characteristics of the GBDT model;

and inputting the output characteristics of the GBDT model into a logistic regression LR model to obtain a recommended value corresponding to each target object.

6. The method of claim 1, wherein the cleansing order data comprises:

identifying the swiped line as a corresponding invalid order;

and screening out the invalid objects corresponding to the invalid orders.

7. The method of claim 1, wherein the cleansing order data comprises:

analyzing order data of a user, and determining a specified object in order placing objects corresponding to the order data, wherein the specified object is an object without a deal amount;

and screening out the specified object.

8. An object recommendation apparatus, characterized in that the apparatus comprises: the device comprises an acquisition module, a first determination module, a second determination module, a third determination module, a sorting module and an output module; wherein the content of the first and second substances,

9. A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements the steps of the object recommendation method according to any one of claims 1 to 7 when executing the computer program.

10. A storage medium having a computer program stored thereon, wherein the computer program is executed by a processor. Implementing the object recommendation method of any one of claims 1 to 7.